Aiming at the problem that the noise reduction effect of fault signal is not ideal by only using SVD or morphology, an SVD noise reduction method based on multiscale morphology was proposed.Firstly, based on the principle of SVD noise reduction, the matrix was reconstructed by using the circulant matrix method.Secondly, considering that the effective order size of the reconstructed signal matrix may affect the denoising effect, and based on the MMRR singular value ratio method, the ratio between the adjacent feature extremums is used as the objective function for estimating the effective order, determining the significant figure of the singular value, and restoring the signal.Finally, an SVD filter based on multi-scale morphology was designed, the appropriate structural elements were selected, and the morphological opening and closing operations were adaptively combined.The simulation results show that the improved method can significantly improve the signal-to-noise ratio, and effectively filter out the impact signal and white noise in the fault signal.Compared with the single SVD or morphological noise reduction method, it has stronger anti-noise ability and has certain engineering application value.